In order to build an effective artificial intelligence (AI) platform, you need good data. Data feeds the algorithms that go into the AI; the better your data the better your AI system will function.
In the food tech world, there are a number of startups like Spoonshot, Analytical Flavor Systems and Tastewise have built intricate AI platforms that use tons of different data to help big CPG companies identify and predict culinary and flavor trends.
But what happens when a big catastrophic black swan event occurs like, oh, I don’t know, a global pandemic, which changes the eating and buying patterns of almost everyone on the planet all at once?
For instance. In February, it was easy to buy flour and yeast at your local grocery store. Fast forward to March and suddenly store shelves were empty and you had to resort to making your own yeast. Around that same time, instead of pictures of fancy restaurant meals, social media accounts were flooded with pictures of homemade bread.
Food predicting AI systems uses data points like restaurant menus, social media mentions and consumer purchasing patterns to determine future trends. But everyone didn’t start making sourdough bread at home because it was suddenly fashionable. It was because everyone was stuck inside.
How then, will AI systems handle this shock to the data system? Sheltering in place won’t last forever (knocks on wood), and who knows how long people will actually make their own bread. The popularity of it now is an aberration, does this mean that the data surrounding it is no good? Is bread making today indicative of anything other being bored or does it foretell a bigger trend?
To get a better sense I reached out to both SpoonShot and Analytical Flavor Systems to see how they are incorporating this massive disruption to our eating patterns into their own prediction process — and got two very different answers.
SpoonShot’s AI uses more than 3,000 sources across 22 data sets including menu, social and pattern data. Kishan Vasani, Co-Founder & CEO of SpoonShot, didn’t seem to think that COVID-19-induced eating changes would impact his company’s predictive capabilities at all. “Algorithms shouldn’t be overly sensitive to black swan data,” he said, “If you think about it, AI essentially means having enough relevant and appropriate data to process and predict.”
In other words, if your AI system is worth its salt, you should be able to weather big changes like this. “Everything goes back to the data and data sources,” Vasani said, “Menu data is significantly slowed down, but that’s compensated for with cooking platforms.”
On the other hand Jason Cohen, Founder and CEO of Analytical Flavor Systems, thinks the pandemic and subsequent lockdowns are a big deal. “Companies will say, ‘no no no, we can make predictions,'” Cohen said, “I do not believe that. This is the most rapid and intense change to consumer behavior since World War II.”
Cohen believes that with quarantines already in place for more than 60 days, new habits will definitely have formed. People will still be baking bread at home. What’s important is to meet this new data where it lives, literally.
Up until the pandemic, Analytical Flavor Systems used a 50 person panel of tasters as part of its data collection. This panel would come into the office to try various on-market foods. But since lockdown, the company has moved entirely to at-home testing. “In addition to CPG products, we are asking them to taste profile their homemade bread and soups,” Cohen said, “The point is we need to see those flavors, aromas and textures they are exposing themselves to.”
Cohen doesn’t think that past data is invalidated, but rather that data needs to be collected before during and after this crisis. Something which I think Vasani would agree with.
The thing about predictions now is that we won’t know if they were accurate for a long time. SpoonShot looks out 18 months and is even considering pushing that out to two years.
Hopefully we’ll be able to eat bread at a restaurant again by then.